A multi-disciplinary data integration method for a remote medical platform

Abstract

This invention discloses a multidisciplinary data integration method for telemedicine platforms, comprising the following steps: S1, the telemedicine platform receives consultation data requests and connects to heterogeneous data sources; S2, a two-layer cleaning model based on a rule engine and AI recommendation is used to deduplicate, correct errors, standardize, and complete missing data from the heterogeneous data sources; S3, during the conversion stage, based on a unified multidisciplinary terminology set and coding library, the data from the heterogeneous data sources are mapped to the same semantic layer, generating a quality score; S4, for large image or pathology files, an intelligent slicing algorithm is used to divide them into blocks <100MB and index them; S5, the text-based structured data and the block file data are uploaded to the telemedicine platform database through different channels; S6, after loading, the multi-source records are merged based on the patient master index to generate a unified multidisciplinary view and write back a quality report. This invention provides reliable data support for consultation experts, improving the efficiency and accuracy of remote diagnosis and treatment.

Description

Technical Field

[0001] This invention relates to the field of medical informatics and data integration technology, specifically to a multidisciplinary data integration method for telemedicine platforms. Background Technology

[0002] With the rapid development of medical alliances and telemedicine, patient data is scattered across multiple heterogeneous systems, including Hospital Information System (HIS), Laboratory Information System (LIS), Imaging System (PACS), Radiology System (RIS), ECG, ultrasound, pathology, and third-party testing centers. This results in problems such as "data silos, semantic inconsistencies, inconsistent quality, and high failure rates in large file transfers." Traditional ETL tools require extensive manual coding of interfaces and conversion scripts, leading to long implementation cycles, high maintenance costs, and an inability to provide real-time assessment and correction of data quality based on remote consultation scenarios. Consequently, they fail to meet the demands of multidisciplinary collaboration for "highly complete, highly consistent, and diagnostically accurate" data. Summary of the Invention

[0003] The purpose of this invention is to provide a multidisciplinary data integration method for telemedicine platforms, which solves the problems of large workload in interface development, uncontrollable data quality, unreliable large file transmission, and inconsistent semantics in the existing technology.

[0004] To achieve the above objectives, the technical solution of the present invention is as follows: A multidisciplinary data integration method for telemedicine platforms, comprising the following steps: S1. The telemedicine platform receives consultation data requests and connects to heterogeneous data sources according to the consultation data requests. S2. A two-layer cleaning model based on rule engine + AI recommendation is adopted to deduplicate, correct errors, standardize and fill in missing data in the raw data of heterogeneous data sources. S3. During the transformation phase, based on a unified multidisciplinary terminology set and coding library, data from heterogeneous data sources are mapped to the same semantic layer, and quality scores are generated. S4. Use an intelligent slicing algorithm to divide large image or pathology files into blocks <100MB and mark them with an index; S5. Upload text-based structured data and segmented file data to the remote medical platform database through different channels; S6. After loading, the multi-source records are merged based on the patient master index to generate a multidisciplinary unified view and write back the quality report.

[0005] As an improvement to the above technical solution, when receiving consultation data requests, the telemedicine platform uses an ETL extraction thread to obtain patient basic information, test results, images, pathology, medical orders, and audio and video data in parallel. The ETL thread extraction formula is: ; Where i represents the type of data source, and Weight i The weights for heterogeneous data sources are given by Q, where Q is the task queue length and N is the number of data sources. The denominator is calculated as the number of CPU cores multiplied by 2 to avoid excessive CPU resource contention.

[0006] As an improvement to the above technical solution, the heterogeneous data sources include: Hospital Information System (HIS), Laboratory System (LIS), Imaging System (PACS), Radiology System (RIS), Electrocardiogram System, and Ultrasound and File System; wherein the weight of Hospital Information System (HIS) is 1.2, the weight of Laboratory System (LIS) is 0.8, and the weight of Imaging System (PACS) is 1.5.

[0007] As an improvement to the above technical solution, step S2 is as follows: S21. The remote client collects the patient's basic information, test results, images, pathology, medical orders, and audio-visual data as raw data. Then, these raw data are classified and stored separately in heterogeneous data sources. S22. The two-layer cleaning model first determines the type of heterogeneous data, dividing it into structured and unstructured data. Structured data is validated using field-level regular expressions, while unstructured data is identified using NLP entity recognition. A predefined rule base is used for comparison, deduplication, error correction, standardization, and missing data removal. Specifically, structured data is validated using field-level regular expressions, while unstructured data is identified using NLP entity recognition. S23. The data cleaned by the two-layer cleaning model is stored in the predefined rule base.

[0008] As an improvement to the above technical solution, the formula for missing information completion is: ; As an improvement to the above technical solution, the semantic conversion process is as follows: Compare the terms from different source data with a standard terminology database to determine whether they are standard terms. If it is a standard term, then it is directly encoded according to the standard terminology database; If it is a non-standard term, it will be matched with a standard terminology database for synonyms. The matching results will be submitted to a human reviewer and then converted into standard terms, which will be mapped to the same semantic layer for storage. The quality of the results stored at the same semantic layer is scored, and the process ends. The standard terminology database includes the SNOMEDCT international standard terminology database and the ICD-10 international standard terminology database.

[0009] As an improvement to the above technical solution, the quality scoring model is as follows: The standardization is the proportion of standard terminology used, with a value range of 0-1; the completeness is the fill rate of required fields.

[0010] As an improvement to the above technical solution, in step S4, the formula for the intelligent slicing algorithm is: Slice size = min(100MB, bandwidth × estimated transmission time / 2); where bandwidth × estimated transmission time represents the theoretical maximum amount of data that can be transmitted under the current network conditions.

[0011] Compared with the prior art, the advantages and positive effects of this invention are: The multidisciplinary data integration method for telemedicine platforms of this invention solves the problems of large interface development workload, uncontrollable data quality, unreliable large file transmission, and semantic inconsistency in existing technologies. Its beneficial effects are: 1. By using a two-layer cleaning process of "rules + AI", data quality is significantly improved and the workload of manual review is reduced; 2. Adaptive slicing and multi-channel parallel uploading ensure stable and efficient transmission of large files such as images and pathology documents; 3. Multi-code fusion mapping and EMPI mechanism enable second-level merging of patient data across institutions and systems; 4. Hot update configuration center, which allows adding new data sources or changing fields without restarting the task, supports rapid expansion of the remote medical platform; 5. Quality scoring and visual feedback provide consulting experts with reliable "diagnostic-level" data support, improving the efficiency and accuracy of remote diagnosis and treatment. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a diagram of the double-layer cleaning model of the present invention; Figure 2 This is a flowchart of the semantic transformation process of the present invention; Figure 3 This is a flowchart of the slicing process of the present invention; Figure 4 This is a flowchart of the data extraction layer of the present invention; Figure 5 This is a flowchart of the data processing layer of the present invention; Figure 6 This is a flowchart of the data loading layer of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention. like Figures 1 to 6 As shown, the multidisciplinary data integration method for telemedicine platforms of the present invention proceeds as follows: 1. Dynamic thread management: an adaptive thread pool algorithm based on multiple data sources; Core principle: 1.1 Input parameters, data source types: HIS (Hospital Information System), LIS (Laboratory System), PACS (Picture Archiving System), etc., 7 types, each corresponding to different I / O characteristics; Number of data sources (N): Dynamically detects the number of currently active connections; Task queue length (Q): The backlog of tasks to be processed; 1.2 Formula for calculating the number of threads: Weight i : Data source weights (HIS=1.2, LIS=0.8, PACS=1.5); Denominator logic: CPU core count × 2 to avoid excessive CPU resource contention; 1.3 Data-driven weight design:

[0015] 1.4 Basis for weight calculation: The response time was weighted at ±0.2 for every ±50ms difference from the baseline (100ms); the coefficients (R²) were determined through regression analysis of historical data. 2 =0.93); 1.5 Dynamic Adjustment Strategy: 1.5.1 Cold Start Phase: initial_threads=min(4,cpu_cores*2) # Conservative initial value; 1.5.2 Runtime feedback control; Monitoring metrics: CPU utilization (>70%, reduce threads), task completion rate (<90%, increase threads); Adjustment cycle: Updated every 30 seconds based on the average of the sliding window; 1.5.3 Extreme Scene Handling: Burst traffic: If the Q growth rate is >10% / second, the number of temporary threads = current × 1.5 (maximum MaxThreads); Deadlock detection: Threads idle for a timeout (default 60 seconds) are automatically recycled; 1.6 Performance verification data: .

[0016] 1.7 Optimization directions: Machine learning prediction: Use LSTM to predict the load in the next 5 minutes and expand capacity in advance; Tiered thread pool: Allocate independent thread groups for critical systems such as HIS / PACS; Containerization adaptation: K8sHPA linkage enables node-level elastic scaling.

[0017] 2. Connect to HIS, LIS, PACS, RIS, ECG, ultrasound, and file system respectively through pre-configured connection pools and drivers; 3. Receive "consultation data requests" from the telemedicine platform and trigger incremental or full data extraction strategies; 4. Utilize each extraction thread to acquire patient basic information, laboratory test results, images, pathology data, medical orders, and audio / video data in parallel; 5. A two-layer cleaning model based on rule engine + AI recommendation is adopted to remove duplicates, correct errors, standardize and fill in missing data in the original data; 5.1 Double-layer cleaning model as follows Figure 1 As shown, Built-in 300+ cleaning rules (such as ID card verification and mobile phone number formatting); Supports custom rule DSL configuration; Rule layer: Pre-defined business rules, regular expressions, and dictionary tables for quickly filtering obvious errors; AI Recommendation Layer: It uses historical cleaned samples to train an anomaly detection model, assigns probability scores to anomalies that the rule layer cannot identify, and recommends corrective values.

[0018] AI recommendation layer (probabilistic completion): ; Predict missing values ​​using similar data context, and output Top 3 completion suggestions for manual confirmation; 5.2 Quality Assessment Indicators:

[0019] 5.3 Data Deepening and Validation: Test data (medical records): Issues with raw data: duplication rate: 12.7%, diagnostic terminology error rate: 8.3%, missing laboratory units: 23.1%; Cleaning results: print(f"Deduplication efficiency: {deduplication_rate:.1%} → Rule engine contributes 82%), print(f"Terminology correction:{correction_accuracy=94.2%}"), AI completion effect display(ai_recommendation.success_rate) # Missing value completion accuracy 89.7%; Key findings: The rule engine is 3 times more efficient than the AI ​​model in correcting errors in structured data (response time <50ms), and the AI ​​recommendation is 41% more accurate than the rule in completing non-standardized fields.

[0020] 5.4 Optimization directions: Dynamic rule orchestration: Automatically selects the rule execution order based on data characteristics; Federated learning reinforcement: Jointly training and completing models across enterprises without sharing raw data; Explainability Reports: Generate visualizations of the decision paths for the cleaning process (e.g., why "heart attack" was changed to "myocardial infarction").

[0021] 6. During the conversion phase, based on a unified multidisciplinary terminology set and coding library, different source data are mapped to the same semantic layer, and quality scores are generated. The unified semantic layer uses a multi-code fusion mapping table based on SNOMEDCT, LOINC, ICD-10, and a custom telemedicine extension code to unify synonyms from different sources for the same patient.

[0022] 6.1 Terminology Set and Encoding Library Construction: Standard Terminology Database: Based on international standards such as SNOMEDCT and ICD-10; Interdisciplinary Mapping Table: Establishes cross-mapping rules such as "laboratory indicators → clinical diagnosis"; Weight Allocation: Sets priorities based on the frequency of term usage (e.g., high-frequency terms = 1.3, low-frequency terms = 0.6).

[0023] 6.2 Semantic conversion process as follows Figure 2 As shown: 6.3 Quality Scoring Model Standardization: The proportion of standard terms used (0-1); Completeness: The completion rate of required fields (e.g., deduct 0.1 points for missing patient gender).

[0024] 6.4 Data Deepening and Validation:

[0025] Key findings: Structured data (LIS) mapping is 15% more efficient than free text (HIS), and human review improves the accuracy of edge term matching from 62% to 88%.

[0026] 6.5 Optimization directions: NLP Enhancement: Using BioBERT to process abbreviations and synonyms in clinical text; Dynamic weight adjustment: The scoring weights are automatically optimized based on departmental characteristics (e.g., surgeons focus more on surgical terminology). Blockchain-based evidence storage: Recording mappings on the blockchain ensures audit traceability.

[0027] 7. For large image / pathology files, an intelligent slicing algorithm is used to divide them into blocks <100MB and mark them with an index; The intelligent slicing algorithm dynamically adjusts the slicing threshold based on network bandwidth, file size, and priority, and only retransmits the corresponding slice when transmission fails.

[0028] 7.1 In-depth analysis of the intelligent slicing algorithm: The transmission threshold formula is slice size = min(100MB, bandwidth × estimated transmission time / 2). This formula optimizes data transmission by dynamically balancing network efficiency and transmission robustness. Formula Principle Layered Analysis: Physical Meaning of Bandwidth × Estimated Transmission Time: Represents the theoretical maximum amount of data (throughput) that can be transmitted under current network conditions. Example: Bandwidth = 10MB / s, Estimated Time = 20 seconds → Throughput = 200MB; Function: Reflects the instantaneous network transmission capacity, serving as a slicing baseline value; Divided by a factor of 2 and core logic: Reserves security redundancy; Network fluctuation buffer: Actual bandwidth often drops by about 30-50% due to congestion / jitter. Divide by 2 to reserve a 50% margin to cope with fluctuations.

[0029] Concurrency control: Avoid a single slice monopolizing bandwidth and ensure the stability of multi-task parallelism.

[0030] Data verification:

[0031] min(100MB) is the engineering constraint: Fault recovery cost: When a 100MB slice fails to transmit, the retransmission time is approximately 5 seconds (10MB / s network); if a 1GB slice fails, the retransmission time is approximately 100 seconds.

[0032] Memory management: The receiver buffer is typically designed to be 100~150MB (Android / iOS best practice).

[0033] Reasonableness of the upper limit: Simulation of retransmission time for slice #1 (bandwidth = 10MB / s); 2slice_sizes=[50,100,200,500]#MB; 3retry_times=[size / 10forsizeinslice_sizes]# seconds; Output #4: [5.0, 10.0, 20.0, 50.0] → Retransmission cost is controllable at 100MB; 7.2 Dynamic Scene Data Validation: Scenario 1: Stable high-bandwidth environment: Parameters: Bandwidth = 50MB / s, Estimated time = 30 seconds; calculate: ; Effect: Triggers the upper limit constraint, preventing the generation of a 750MB large slice; Slice transfer time = 2 seconds (<< estimated time), supports high-concurrency slices; Scenario 2: Weak network fluctuation environment: Parameters: Bandwidth = 2MB / s, Estimated time = 60 seconds; calculate: ; Effect: Theoretical transmission time = 30 seconds, actual tolerance time after reserving 50% redundancy = 60 seconds; Even when the bandwidth drops to 1MB / s, the transmission can still be completed within 60 seconds; Scenario 3: Extremely low bandwidth: Parameters: Bandwidth = 0.5MB / s, Estimated time = 120 seconds; calculate: ; Robustness verification: If the network is intermittently interrupted, the cost of retransmitting a 30MB slice is only 1 minute (0.5MB / s). Compared to the default 100MB slice without an algorithm: retransmission takes 200 seconds, and the risk of timeout increases by 300%; 7.3 Algorithm advantages and engineering value; Adaptive network state: Dynamically adjust slice granularity: approach the 100MB upper limit when bandwidth is sufficient, and automatically reduce slice size when the network is weak; Actual test comparison (1GB file transfer):

[0034] Optimal utilization of resources: Avoid the proliferation of small slices: Slices larger than 10MB can reduce the TCP header overhead to less than 2% (compared to 1MB slices accounting for approximately 15%). Smooth memory usage: Peak receive-end buffer size is controlled within 100MB; Fault tolerance and retransmission optimization: If a slice fails, only a maximum of 100MB of data needs to be retransmitted. Combined with incremental verification: only failed shards are verified (Hash comparison), reducing verification overhead by 30%; Implementing the enhancement strategy: Key enhancements: Bandwidth probing: Dynamically calibrate real-time bandwidth via TCPRTT (Round-Trip Time); Estimated time correction: based on a weighted average of historical transmission success rates (e.g., exponential smoothing). Concurrency control: Number of slices ≤ Bandwidth / (Slice size / Estimated time); This intelligent slicing algorithm achieves an optimal balance between transmission efficiency and reliability through a three-layer design: transmission capacity assessment, security redundancy reservation, and engineering constraint protection. In the formula, bandwidth × time reflects the network's physical limits, ÷ 2 represents the core robust strategy for handling fluctuations, and the 100MB upper limit is the optimal solution for engineering practice. Practical deployment recommendations: Combine the QUIC protocol to reduce the impact of head-of-line congestion, and enable FEC (forward error correction) in weak network environments to further reduce the retransmission rate.

[0035] 8. Upload text-based structured data and segmented file data to the remote medical platform database through different channels; 9. After loading is complete, multi-source records are merged based on the patient master index (EMPI) to generate a multidisciplinary unified view and write back the quality report.

[0036] Based on the embodiments of the present invention, any modifications, equivalent substitutions, improvements, etc., made by all other embodiments obtained by those skilled in the art without creative effort should be included within the protection scope of the present invention.

Claims

1. A multidisciplinary data integration method for telemedicine platforms, characterized in that: The steps of this method are: S1. The telemedicine platform receives consultation data requests and connects to heterogeneous data sources according to the consultation data requests. S2. A two-layer cleaning model based on rule engine + AI recommendation is adopted to deduplicate, correct errors, standardize and fill in missing data in the raw data of heterogeneous data sources. S3. During the transformation phase, based on a unified multidisciplinary terminology set and coding library, data from heterogeneous data sources are mapped to the same semantic layer, and quality scores are generated. S4. Use an intelligent slicing algorithm to divide large image or pathology files into blocks <100MB and mark them with an index; S5. Upload text-based structured data and segmented file data to the remote medical platform database through different channels; S6. After loading, the multi-source records are merged based on the patient master index to generate a multidisciplinary unified view and write back the quality report.

2. The multidisciplinary data integration method for a telemedicine platform according to claim 1, characterized in that: When receiving consultation data requests, the telemedicine platform uses an ETL extraction thread to obtain patient basic information, test results, images, pathology, medical orders, and audio and video data in parallel. The ETL thread extraction formula is: ; Where i represents the type of data source, and Weight i The weights for heterogeneous data sources are given by Q, where Q is the task queue length and N is the number of data sources. The denominator is calculated as the number of CPU cores multiplied by 2 to avoid excessive CPU resource contention.

3. The multidisciplinary data integration method for a telemedicine platform according to claim 1, characterized in that: The heterogeneous data sources include: Hospital Information System (HIS), Laboratory Information System (LIS), Imaging System (PACS), Radiology System (RIS), Electrocardiogram System, and Ultrasound and Documentation System; among which, the weight of Hospital Information System (HIS) is 1.2, the weight of Laboratory Information System (LIS) is 0.8, and the weight of Imaging System (PACS) is 1.

5.

4. The multidisciplinary data integration method for a telemedicine platform according to claim 2, characterized in that: The steps in step S2 are as follows: S21. The remote client collects the patient's basic information, test results, images, pathology, medical orders, and audio-visual data as raw data. Then, these raw data are classified and stored separately in heterogeneous data sources. S22. The two-layer cleaning model first determines the type of heterogeneous data, dividing it into structured data and unstructured data. Structured data is validated using field-level regular expressions, while unstructured data is identified using NLP entity recognition. A predefined rule base is used for comparison, deduplication, error correction, standardization, and missing data removal. Structured data is validated using field-level regular expressions, while unstructured data is validated using NLP entity recognition methods. S23. The data cleaned by the two-layer cleaning model is stored in the predefined rule base.

5. The multidisciplinary data integration method for a telemedicine platform according to claim 1, characterized in that: The formula for missing completion is: .

6. The multidisciplinary data integration method for a telemedicine platform according to claim 1, characterized in that: In step S3, the semantic conversion process is as follows: Compare the terms from different source data with a standard terminology database to determine whether they are standard terms. If it is a standard term, then it is directly encoded according to the standard terminology database; If it is a non-standard term, it will be matched with a standard terminology database for synonyms. The matching results will be submitted to a human reviewer and then converted into standard terms, which will be mapped to the same semantic layer for storage. The quality of the results stored at the same semantic layer is scored, and the process ends. The standard terminology database includes the SNOMEDCT international standard terminology database and the ICD-10 international standard terminology database.

7. The multidisciplinary data integration method for a telemedicine platform according to claim 1, characterized in that: The quality scoring model is as follows: The standardization is the proportion of standard terminology used, with a value range of 0-1; the completeness is the fill rate of required fields.

8. The multidisciplinary data integration method for a telemedicine platform according to claim 1, characterized in that: In step S4, the formula for the intelligent slicing algorithm is: Slice size = min(100MB, bandwidth × estimated transmission time / 2); where bandwidth × estimated transmission time represents the theoretical maximum amount of data that can be transmitted under the current network conditions.

Images